It’s an average and is well-known in polling statistics and there are statistical adjustments that can help to a degree but not when the non-response rate is as high as it is in political polling.
Every survey in statistics is scientifically valid if the random sample is representatively drawn from the population. It doesn’t mean the survey will be true or even accurate. It merely suggests that if the underlying sample is representative, then repeated surveys would have an accurate average.
Every survey can bring out an outlier or luck of the draw even if the sample is validly drawn.
Political polls are so noisy, so shoddy that the only people doing the analyses are considered flunkies because they are no better than liars, their ethics would never be accepted in reputable statistical journals. They are so debauched that they are not even allowed to be called statisticians, rather they are called pollsters. And they do it for one reason only; money.
For example, their political polls will take a designed sample plan and ruin it with non-response, yet they persist in doing the analysis even though the results are invalid on their face. They do not account for the non-responses and they never disclose these rates. They base their MOE on ‘responses’ only, which only adds insult to injury. If they disclosed the non-response rates, the MOE would not be +/-3% or even +/-6%. With the high non-reponse , the MOE would be more like +/-30%.
It is insulting to see certain polling outfits pat themselves on the back in self-congratulations for getting the closest prediction to an actual outcome. It is nothing more than luck of the draw.
If I play a hand of poker and lay down a Royal Flush, it does not mean I am a champion. It is only after many many observations of playing hands where my caliber as a player can accurately be estimated.
In the arena of political polling, there are no champions. Such an arena is a circus for flunkies who could never make it through a rigorous statistical curriculum. They do it for pay and because they have enough skill to do a spreadsheet. But they have no talent to jackknife, bootstrap or stratify sampling, etc. And if they should have such understanding, and they still participate in such a liar’s field, then it is because they have character defects.
The 90% is an average that has been talked about in statistical journals and journals that use statistics heavily. True and reputable statisticians have been requested innumerable times to design methodology to adjust for non-response rates. It can only be done when the rate is not too high.
http://link.springer.com/chapter/10.1007%2F978-1-4020-5666-6_8
There are hundreds of publications on the subject. No one can raise dead data to life, that is what it boils down to. Look at these search terms on the internet and you will find numerous references to high non-response rates in political surveys. I am looking at a list of 20 on the first page of a search.
There is no statistical method in the realm of the possible that can project an accurate unbiased estimate when the underlying random sample has been destroyed. It is no more possible to do that than it is to make a gas engine run on a fuel tank filled with water.
The political candidates use polls for cheerleading and for sales pitches. Although all the polls are invalid via invalid samples, there are two broad categories of invalid samples. One category is invalid because targeted persons refuse to participate and the other is where the targeted people are oversampled in a set of attributes. Wikileaks has revealed the democrats and Clinton are deliberately pushing the latter while Trump is subject to the former.
There is one valid aspect to the samples that are obtained for Trump, that is from the frustrated underemployed or dissatisfied American worker sitting at home. With 90 million people out of work or underemployed, the respondents are expressing more support for Trump because of their predicament. Even as the nonresponse rates are too high for a valid estimate, the invalid sample picks up on those at home that are frustrated and dissatisfied. But overall, the statistics in this arena are rubbish.
I appreciate all the time you took to write up that reply. As an engineer, I appreciate statistical processes. I can see where the whole random sample thing breaks down. I can also see while “pollsters” would want to hide the percentage of non-response.
My gut feeling is that the folks who are actually working 8-10 hours a day, or those that wish they were, are mad enough that they would tell pollsters to eff-off, and those parasites who live at home on un-deserved govt freebies are the type who would reply so as to support further mooching.